A new study tested ChatGPT, Gemini, Grok and Claude with 100 politically charged questions — and found every major AI chatbot fell for falsehoods from both sides of the aisle while citing sources that literally don't exist.
Of 419 sources the four chatbots provided, 104 webpages didn't exist, and 77 didn't actually answer the questions posed. Sources were valid less than half the time overall.
The study, conducted by Just Facts, found ChatGPT and Gemini performed at a "C student" level, Claude at a "B student" level, and Grok scored better on left-leaning falsehoods but worse on the right — the reverse pattern of the other three. The researcher behind it says the takeaway isn't to abandon AI, but to never blindly trust it: "Don't trust, verify."
Of 419 sources the four chatbots provided, 104 webpages didn't exist, and 77 didn't actually answer the questions posed. Sources were valid less than half the time overall.
The study, conducted by Just Facts, found ChatGPT and Gemini performed at a "C student" level, Claude at a "B student" level, and Grok scored better on left-leaning falsehoods but worse on the right — the reverse pattern of the other three. The researcher behind it says the takeaway isn't to abandon AI, but to never blindly trust it: "Don't trust, verify."
A new study tested ChatGPT, Gemini, Grok and Claude with 100 politically charged questions — and found every major AI chatbot fell for falsehoods from both sides of the aisle while citing sources that literally don't exist.
Of 419 sources the four chatbots provided, 104 webpages didn't exist, and 77 didn't actually answer the questions posed. Sources were valid less than half the time overall.
The study, conducted by Just Facts, found ChatGPT and Gemini performed at a "C student" level, Claude at a "B student" level, and Grok scored better on left-leaning falsehoods but worse on the right — the reverse pattern of the other three. The researcher behind it says the takeaway isn't to abandon AI, but to never blindly trust it: "Don't trust, verify."
0 Comments
0 Shares
2 Views
0 Reviews
العربية